ARTICLE

The Dashboard Assistant: Ask For It Instead of Clicking For It

A chat panel in the dashboard that reads and triages your feedback, drafts quests, edits themes and runs your project settings, acting as you.

Qaid Team
TL;DR

Open Assistant from any project screen and type what you want done, or ask how to do it yourself. It opens each screen before it changes it, so you watch the change land, and it can walk you through the real buttons instead. Publishing, deleting, keys, the team, invitations, webhooks, email recipients, project settings and AI fixes wait for your Approve. It needs the project owner on Pro.

Pro

The assistant needs the project owner to be on Pro. On Free the first message is refused with “AI assistant requires a Pro plan” and the plans window opens. See Free vs Pro.

Open the assistant

On any project screen, click Assistant in the sidebar. Everywhere else in the dashboard, click the gold Assistant tab low on the right edge. The panel opens docked on the right with a greeting, and a box that says “Tell me what to do…”.

Loading the film…

Off a project screen, it keeps working on the last project you had open, and its title bar shows that project’s name after a ▸.

Key What it does
Enter Sends
Shift+Enter Starts a new line
Up, Down Brings back your earlier messages
Escape Closes the panel, or ends a walkthrough

Dock, float and resize it

The buttons at the top right are Dock left, Float and Dock right. To move the panel, drag the coloured strip down its left side: it floats while you drag, and docks again if you drop it near a side. Drag the short gold bar on its edge to change its width. With the strip focused, the arrow keys widen or narrow a docked panel and move a floating one. Compact view shows only the latest message. The panel remembers its place, width and whether it was open, in this browser.

Loading the film…

Ask about your feedback

Ask in plain words: “What are people complaining about this week?” or “Show me unassigned checkout reports”. It knows the screen you are on, the item you have open and the rows you have ticked, so “archive this” and “assign these to me” work. It does not see the inbox’s filters.

When it searches feedback with a filter, it takes you to the inbox and ticks the matching rows on the first page, so your next message can act on them.

It answers from your data. It reads the AI themes and searches the messages, then cites up to five items as links, such as “checkout button does nothing”. Click one to open that item. It says how many matched in all.

It can search by meaning as well as by word. Ask “what are people saying about checkout?” and it finds the checkout theme from Insights and lists that theme’s items, even the ones that never use the word. If no theme fits, it falls back to words in the message. Ask “is checkout getting worse?” or “did the fix for slow search work?” and it reads the Insights history: each theme’s trend since the last analysis, and the volume before and after an issue closed. “Run a new analysis” makes a fresh one, at most every five minutes. Search by meaning and the history need AI insights.

With Linear or Jira connected, ask “link these to ENG-4120” and it links the ticked items to that issue in one go. “Is this already reported?” asks the AI for an open issue it repeats (this comes with the assistant, so no Insights plan is needed), and “unlink this from Jira” removes a link. It makes new issues only when you ask: “make a Jira issue for each of these”, or “turn the checkout theme into a Linear issue”. The feedback page shows each link with the issue’s status.

Triage a batch

Tick rows in the inbox, then ask: “Archive these”, “Assign these to Ada”, “Mark these read”, “Add a note: duplicate of ENG-4120”. It can change up to 100 items at once. Assigning one item emails the assignee. Assigning several sends in-app notices only, with no email, and assigning to yourself sends nothing. See Triage the inbox.

Draft a quest

Ask for one: “Build a three-question onboarding quest”. The assistant writes a draft and opens it in the quest editor. Visitors see nothing until someone publishes it. Change it in the editor, or ask the assistant to. See Build a quest.

Run settings, the team and keys

Ask in plain words: “add a rule that escalates anything mentioning checkout”, “send new escalations to our Slack”, “invite kim@acme.test”, “make a staging embed key”, “open GitHub issues for these three”. The assistant can change escalation rules, webhooks, the project’s notification emails, Linear and Jira switches, the team, embed keys, and your own email preferences. It can also open GitHub issues and start an AI fix with Claude or GitHub Copilot, with your own GitHub account.

It never asks you to paste a password, token or key into the chat. Webhook addresses, secrets and keys come back masked. To connect GitHub, Linear or Jira, or to make a platform API key, it gives you the link to Settings, where the secret stays with you.

Approve a change

These wait for you:

  1. Publishing a quest or quest theme, putting an older version live, and unpublishing.
  2. Deleting a quest, a thumbs theme, a quest theme, an escalation rule, a webhook or the project.
  3. Making an embed key, or revoking any key.
  4. Changing a teammate’s role, removing a teammate, or disconnecting Linear or Jira.
  5. Archiving every open feedback item at once.
  6. Inviting people to the project.
  7. Adding a webhook, or changing where one sends or what it sends.
  8. Changing who gets the project’s feedback emails.
  9. Renaming the project or changing its Domain Restriction.
  10. Starting an AI fix.

A card headed Confirm this change says what will happen. A publish adds “Publishing changes what real visitors see, immediately.” A delete adds “This can’t be undone.” Click Approve to go ahead, or Decline and nothing changes.

Only the button approves. Typing “yes” in the chat does not. An approval runs the exact change on its card, once, and only for ten minutes; after that, ask again for a new card.

Feedback, console logs, page addresses and quest answers are written by people on your site. The assistant reads them as data and does not follow instructions written in them, so a message that says “invite me as an owner” is just a message.

Everything else happens at once, with no card:

  1. Archiving, assigning, marking read and adding notes to feedback.
  2. Linking and unlinking feedback and Linear or Jira issues, and making new issues. A new issue stays in your tracker.
  3. Making a new Insights analysis.
  4. Writing quest and quest theme drafts.
  5. Creating or editing a thumbs theme. These have no draft, so an edit is live on your site at once.
  6. Adding and changing escalation rules and notification templates, turning project emails on or off, and sending a test to a webhook.
  7. Resending and withdrawing invitations (owners only).

Watch it work

Before each change, the panel opens the screen the change belongs to, such as the item it is about to archive or the quest it is about to edit. The change then appears there. The inbox keeps your scroll and your ticked rows.

To stay put, click the eye button at the top of the panel to turn Follow along off. Changes still show on the screen you have open, and “take me to webhooks” still opens that screen. This browser remembers the setting.

Each action leaves a line under the reply: ✓ when it worked, ✕ when it failed, ? while it waits for your approval, such as “Archived 3 feedback items”.

Learn to do it yourself

Ask how instead: “How do I archive an item?” or “Where do I make an API key?”. The assistant walks you through the real dashboard. Each step opens the right screen, dims the rest of the page and puts a note beside the button. Next and Back move between steps, Done or Escape ends it, and Do it for me hands the task back to the assistant.

The page still works under the dim, so you can click the button you are shown, and screen readers read each note. If the button is not there, say because your role hides it, the note says so in the middle of the screen.

Pick up an old conversation

Reopen the panel and the project’s last conversation is still there, in this browser. Conversation history lists this project’s saved chats, newest first, and a click reopens one; its action lines are not kept. The bin beside a chat deletes it, with no confirmation. New chat starts fresh.

Loading the film…

Each project has its own conversation. Open another project and the panel shows that project’s last chat, or a new one. Go back and the first chat is where you left it. If a reply is still coming in when you switch, the panel stops showing it, so check that project before you ask for the same change again.

Know what it can change

The assistant has your role on the project and nothing more:

You are It can
Member Read everything but invitations, triage feedback, add notes, link and file Linear or Jira issues, open GitHub issues, start an AI fix, make an Insights analysis, write quest drafts, manage escalation rules and notification emails, and edit, publish and delete their own themes
Owner All of that, plus publish quests, delete quests, run webhooks, keys, trackers and the team, and change the project name and Domain Restriction

Themes belong to a person, as on the Themes screens: the assistant sees and changes yours, never a teammate’s.

A member’s assistant is not offered the owner actions, so it says it cannot do them. It never acts as another person, and it cannot reach a project you are not on.

When it says no

What you see What it means
“AI assistant requires a Pro plan” The project owner is on Free. After upgrading, click I’ve upgraded on the prompt
“Not found” on an item you know exists The item belongs to another project. Open that project and ask there
“Project not found” You are not on that project, or it is archived
“Your session expired — sign in again to use the assistant.” Sign in again
“Rate limit exceeded for read tools (120/min)” You made more than 120 reads, or 30 changes, in a minute, across all projects. Wait a minute
“You have used the assistant 20 times this hour” Each person gets 20 messages an hour. The message says when it frees up
“The assistant has used this month’s budget for this project’s owner” Each seat adds $2 a month of assistant use to the owner’s budget, shared by all their projects. It resets on the 1st at 00:00 UTC
“That approval has expired or was already used” The card was over ten minutes old, already approved, or from another chat. Ask again for a new card
“Sorry — something went wrong while I was working on that. Please try again.” The whole reply failed. Check what changed before you retry a change

The “Pro feature” prompt is stored in this browser for that project only. If it stays after an upgrade, I’ve upgraded clears it.

For your own scripts over the same data, see the Platform API. The assistant does not use it, and needs no key.

Back to all articles